电子测试
電子測試
전자측시
ELECTRONIC TEST
2013年
20期
32-34
,共3页
赵余%李百川%李鑫%罗万春
趙餘%李百川%李鑫%囉萬春
조여%리백천%리흠%라만춘
人脸朝向识别%随机向量%自组织竞争网络
人臉朝嚮識彆%隨機嚮量%自組織競爭網絡
인검조향식별%수궤향량%자조직경쟁망락
face toward recognition%random vector%self-organizing competitive network
像素可反映人脸特征，通过随机权重可将像素矩阵变换为一个N维向量，从而可用自组织竞争网络模型进行识别。随机产生一系列权重向量，可得当N为10和14时识别正确率最高，达到92%。再分别产生100个随机权重向量，并作识别模拟，平均正确率分别为91.62%和88.56%，标准差分别为0.0194和0.0148。模型识别率高、稳定性好，是一种有效的人脸朝向识别模型。
像素可反映人臉特徵，通過隨機權重可將像素矩陣變換為一箇N維嚮量，從而可用自組織競爭網絡模型進行識彆。隨機產生一繫列權重嚮量，可得噹N為10和14時識彆正確率最高，達到92%。再分彆產生100箇隨機權重嚮量，併作識彆模擬，平均正確率分彆為91.62%和88.56%，標準差分彆為0.0194和0.0148。模型識彆率高、穩定性好，是一種有效的人臉朝嚮識彆模型。
상소가반영인검특정，통과수궤권중가장상소구진변환위일개N유향량，종이가용자조직경쟁망락모형진행식별。수궤산생일계렬권중향량，가득당N위10화14시식별정학솔최고，체도92%。재분별산생100개수궤권중향량，병작식별모의，평균정학솔분별위91.62%화88.56%，표준차분별위0.0194화0.0148。모형식별솔고、은정성호，시일충유효적인검조향식별모형。
The pixel can reflect facial feature,and matrix of pixels is converted to an N-dimensional vec-tor by random weight,which can be used to identify the self-organizing competitive network model.Randomly generate a series of weight vector,the correct rate of identify when N is 10 and 14 is the highest,which is reached 92%.Then generate 100 random weight vector for identification simulation,the average accuracy rate is 91.62% and 88.56% and the standard deviation is 0.0194 and 0.0148.The model is of high rate of identifi-cation and good stability,which is an effective model to recognize face toward.
